speedtables startup locking sometimes fails to provide mutual exclusion / fails outright about speedtables HOT 4 CLOSED

It looks like lockfile is fundamentally broken. It assumes that tcl's "file rename" without -force is atomic, i.e. it will either rename the file without clobbering, or fail.

This is not true. It first checks if the target exists. If the target does not exist, then it does a rename() call, which atomically replaces the target if it exists. There is a window between the test and the rename where another process could create the lockfile.

I'd suggest just using TclX's "flock" (open the lockfile in a+ mode, try to flock -nowait, keep the lockfile open until you want to release the lock)

from speedtables.

or you could probably do something with opening the lockfile with flags CREAT EXCL

from speedtables.

I've created a "locking" branch off "stapi-test" to work on this. Creating the lockfile directly creates a potential race condition where the lockfile is created but the PID of the opener hasn't been put into it. I think I've handled that. Can you give it a try?

from speedtables.

Fixed in 1.6.4 release.

from speedtables.